Pilot Performance With Predictive System Status Information
Anna C. Trujillo
NASA Langley Research Cbnter
Hampton, VA 23681-0001
ABSTRACT
Research has shown a strong pilot preferencE for predictive
information of a iicraft system status in the flight cfeck.
However, the beneiits cf predictive information have not
been quantitatively demonstmted. The study described here
attempted to identity aid quantify these benefits if they
existed. I n his amulator ecperiment, three types of
predictive i iformation (none, whether a patameta" was
changing abncmially, and the time for a parameter to leach
an alert lan^) and four initial times to ai alat (1 minute, 5
minutes, 15 minutes, and ErA+45 ninutes) were found to
affect when subjects axompHshed cErtain a;tions, such as
accESsing p eitinent checklists, cfeclaring anergaicies,
diverting, and calling the fight attaidant and dispatch.
INTRODUCTION
Much a necdotd evidence exists i^arding the benefits
predicting aircraff system fiiluies would bring to increasing
the safety of fli^t. Ebcumaited instances exist where some
type of early notification to the fli^t crew of a system
parameto" deviation could have paevented or lessened the
consequaices of an airaaft's system fiilure [1], [2]. The
benefits of predictive infomiatian, some have argued, are in
the realm of improved (tcision making [3]-[5]. Thus, to
describe and quantify tie benefits of predictive information,
a Esearch program was undeitaken to ^stematically ecplore
these potential benefits.
Eadier research has shown a strong pilot preference for
predictive information [6] aid for certain types of predictive
information, specifically, whether a ^stem parameter (e.g.,
oil t emperature) vias changing abnormally and the time
remaining until that parameter reached a predefined \alue
[7]. Pilots also indicated when they wanted to be told of a
possible problem. For ecample, pilots only wanted to be
notified that the parameter was moving if it would reach ai
alfft range h less than 5 ninutes; otherwise, tiey would
rather know tie approximate time remaining until the alert
[7].
Objectives
There w ere several objectives of this ecpeiiment The
primary objective was to idaitify the benefits cf predictive
information in an cperational setting during non-normal
system events. Benefits were defined £B decision making
pertaining to handling of the ladt, vhich included taking
actions to aifect the onset of an alert, retrieving dieckUsts,
diverting a nd declaring aneigencies, aid wcrkload
associated w ith the non-normal system events. The
seoondaiy objective was to cbtermine tie most appropriate
form o f predictive information and range of prediction
times. Lastly, this experiment was conducted to corroborate
previous research.
Experimaital Variables
Of the t hree etperimental variables, kvo were directly
manipulated: the predictive information available and the
inrtial time to an alert. The predictive information available,
a betweai subject variable, was cne of three types: (1) none
(baseline), ( 2) whether a parameter was inaeasing or
decreasing abnormally direction), or (3) he time to an dert
(time). The initial time to an alert, a within subject variable,
had four levds: (1) 1 minute, (2) 5 ninutes, (3) 15 minutes,
and (4) ErA+45 minutes (Estimated Time to Arival).
Baseline predictive information aid ETA+45 minute initial
time t o an alert were control conditions. The third
experimental variable, which was partially controlled in hat
the parameter would degrade in aregulated mannff, was the
four independent hults each subject aicountered
Predictive Information: In the baseline condition, no
predictive i rformation was a/ail^le. Thus, when a
parameto" rexhedan alert range, the subjects saw the typical
alert message (e.g, CABIN ALT) with he axompanying
auial alert (able 1).
Table 1 -Examples cf Predictive Information
Condition
Baseline
Directicm
Time
Predictive Information
"CABIN ALT INC"
"CABIN ALT 7MIN
Alert Information
"CABIN ALT"
"CABIN ALT"
"CABIN ALT"
In the other two conditions, direction and time, subjects
were notified that a parameter was moving towards an alert
range. In all cases, the predictive information presented to
subjects was always correct and had an alert category of
advisory. Furthermore, parameters increased or cfecreased a a
constant rate dependent on he state cf the airaafl. Lastly,
whai t he parameter reached ai alfft range, the lelated
standard alert information messa^ replaced the predictive
information messa^ (t^le 1).
Fort he direction condition, subjects were told that a
paiametff was increasing or decreasing abncamally (t^le 1).
For the time condition, subjects were told when a paiametff
would reach ai alfft range fir the given aircrafi state (table
1). The lime to an alert was updated in increments of whole
minutes if the time remaining was greater tian 1 minute. I
the time to ai alat was less than 60 seconds, tie message
updated for a'ery 15 -second change in the ime to an alert
Initial Time to an Alert: Each subject saw four initial
times to an dert (the time to an aleit at the beginning cf a
failure): (1) 1 minute, (2) 5 minutes, (3) 15 minutes, and
(4) E TA+45 minutes. The configuration of the aircraft
affected the xtual time to an alert; for ecample, throttling
back the aigine with the EGT (Exhaust Gas Temperature)
inaease would increase the time to an alert.
Faults: Ea;h of the four data runs, or s:enarios, and
the training run included afault in \^hich a parameta" would
eventually reach ai alert range f the subject taok m xtion.
The iaults wa"e (1) cabin altitude inaease, (2) forward cargo
overheat, (3) EGT increase, and (1) oil quantity decrease.
The training run was an avicxiics overheat. All fiilures were
designed to behave as realistically as possible [8]-[ll] and
are described below.
For the scenario wth tie cabin dtitude increase, the cabin
altitude i ncreased to tie airplane altitude The cxitilow
valve, ff (iiecked, was in the folly dosed condition once the
failure started. A Ithough the inaease could not be
controlled t Irough the mvironmental ^stem, the cabin
altitude w aming wDuld not be reached if the subject
descended to no more than 10 000 ft and if he had at least
4_ minutes until the alert range was to be reached — the
time needed to descend fi-om the initial altitude of 37 000 fi
to 10 000 ft.
In another data run, the forward cargo hold, initially set-up
for carrying animals, had a tempaature inaease until it
rexhed the fre \^aming limit. I the subject (hanged to the
cargo mode, the temperature increase would slow. Also, if
he discharged the forward cargo fre bottle befcre the alot
range was reached, the forward cargo temperature would
never reach the alert range, ff tie subject disdiarged the
forward cargo lire bottle after tie fire warning, as the
forward cargo fire diedilist instructs him to do, the
temperature W3uld drop below the alert range.
During the scenario with the EGT increase, the HjT rose
steadily and if it readied tie alert range, the subject would
have to follow the engine iailure'shutdown procedure. The
inaease could be slowed if the subjed throttled bade the
engine with the increasing HjT or stopped ff the affected
engine was shut dowoi. I the subject restarted the oigine,
the E GT would ^ain inaease until it reached the alert
range.
The scenario with the dl quantity deaease also involved an
oil pressure decrease because of the bss d" dl. The oil
pressure tri^ered the alert once it reached an alert range.
The only way to decrease the rate of dl loss was to shut
down the affected engine.
For t he avionics cverheat training run, the rate of
temperature increase could be deaeased by dianging tie
avionics mode t o werride fiom its initial position rf"
normal. Furthermore, b y disconneding bus 3, the
temperature would stay below thewaming Umit. Thus, the
load on bus 3 was the pimary cause of the overheat.
EXPERIMENT DESIGN
Subjects
Twdve glass-cockpit airline pilots iamiliar with ETOPS
(Extended Twin engine Q'erationS) rules participated as
subjects. S even wae currently first dBcas with the
remaining five captains. The average age was 48 years old
and the averse commerdal airline flight ecperience was 16
years.
Test Design
The expoiment was run in the Advanced Civil Transport
Simulator at the N\SA langley Research Center. This
simulatcr had fli^t paformance characteristics similar to a
Bodng 757. The flight deck resembled a Bodng 71-7400
orMD-11. The subject acted as captain, pilot -not-flying. A
confederate first officer (F/0) was pilot-flying and he was
well versed ii flie qieration of the simulator. A confederate
air traffic controller (ATC) and company dispatdi opaator
provided the necessary coordination with the ground.
The flight was fan Dulles arport to Charles de Guafle
airport with a 604ninute ETOPS rule; ie., the jiane was
never more than 60 minutes fom an altemate airfield The
604ninute rule was used in ader to have several PETs
(Point of Equal Time); i.e., the point where the plane was
60 minutes f rom any suitable altonate airport. The
scenarios were set-up sach that ach sgment of flight
started before a FET; tius, this experiment only included
the cruise phase d flight. If the configuration of tie aircraft
did n ot diange during tie fiult, the affected parameter
would reach ai alert range a few minutes before the arcrafl
interseded t he PET except in the ETA+45 minute
condition.
Any m aterials and information the subject needed were
provided to lim. Hotting diarts, landing jlates, a dspatch
weather briefing, and a flight plan wae available in papa
form. Checklists were dectronic and nimicked the Bodng
model o f the quick refaence handbook [12]. Voice
communication was used for ATC and dispatch Both ATC
and dispatch were able to s ipply current weather
information at any of tie diversion airfields. Basically, the
weather at a 11 diversiai airfields was xcqitable for
landing — driz^e w ith a ceiling around 1 000 ft aid
visibility approximately 1_ miles with wnds at no more
than 10 knots. ATC also reasonably ecpedited any requests
subjects had regarding course changes. The confederate F/0
was able to aiswer operational questions fiom the subject;
i.e, he supplied all lie operational information normally
found in the aircraft manual. Lastly, subjects nade aiy
passenga announcements or Md (onferences with the head
flight atendant, cr purser, to tie experimenter sitting in the
bade of the amulator.
As mentioned earlier, tie iaults and initial times to an dert
were w ithin-subjed variables while the predictive
information was bdween subjects. Since subjects could
only see each lailure mce, each subject had four data runs in
addition to a training run. Thus, the overall results is hat
all subjects saw ®ch d" the four faults once and eadi of the
four initial times to ai alert once with one of the tiree types
of predictive information.
Dependent Measures
The depaident measures consisted of variables that defined
whether the predictive information was beneficial: when and
whffe (Htain actions occuned, aid workload ratings, whidi
were measured u sing the NASA-TLX c|uestionnaire oi
perceived workload [13]. VariEbles not directly dq^endent
on a particular iailure were whai the subject limed off-
track, diverted to an ETOPS alternate airport, bought up
the appropriate checklist, aid initiated action pertaining to
it; the time and ^^ace definition of tie aircrafi; and the
workload ratings. Variables that were directly (tpendent oi
the failure iavolved actions the subject could teke to affect
the time to ai alert, such se when an engine was shut dovm
for the EGT increase scenario and the oil quantity decrease
scenario.
Procedure
Whai a subject first airived he received ai overview on this
experimait including i rstructions about the NASA-TLX
questionnaire After this introduction, the confederate F/0
gave a d etailed description of the simulator and its
operation, and the ilight plan to the subject before the
training run started. The tiaining run included the a'ionics
overheat fault 1 5 minutes into the flight. The ime to an
alat was 5 minutes given the initial aircraft cnnfiguration.
No data were recorded diring training.
A short break w as taken after the training run md before
data run 1. Ai hour lunch beak fallowed the first data nm.
After lunch, the subject completed data runs 2 through 4.
Each data run took approximately 30 minutes. At the aid
of each data run, the subject was asked about the iailure, his
actions, a nd his workload. The presentation order of
predictive i rformation and iiitial time to an alert were
counterbalanced while scenario order was only partially
balanced due to the number cf subjects.
Data Analysis
For time data, arormalized time was alculated to extricate
the fact that different initial imes to an alert occurred
during he flight, ff Jhe times were rot normalized, the Ata
clustered around f)ur dscrete categories cbpendent en the
initial time to an alert. The normalized time was
normalized time -■
time at which X occurred
actual time to alert
Times were Idcen fom iiluie start. The actual time to alert
was when the alert truly occurred or would have occurred
had the subject not done something to p"evait it such as
shut down an engine. These imes were then analyzed iBing
the ^neral inear modd in SPSS [14].
The specific acticms analyzed were accessing the appropriate
checklist, turning o ff path, divating, dedaring an
emffgency, checking the weather a the diversion airports,
calling t he fight attendant, and calling dspatch.
Categorical data lelated to these actions were aialyzed with
the i ndqjendent samples Chi-squared ^)test in SPSS
[14].
All failures had a checklist associated with than. Thus, ff a
parameter reached an alert, the aabject should fellow the
checklist. A subject could access the checklists befcs"e the
alert range was reached if he so desired.
Under ETOPS rules, subjects had to divert fir the oil
quantity decrease and EGT increase Mures when they shut
down an engine. T he ETOPS rules d) not spedfy a
diversion is necessary with cabin pressure loss, but for fiel
efficiency reasons and passaiger comfort, Jhe logical choice
would be to dvert ff a subject had a forward (srgo fire
warning, he would have to divert under ETOPS rules. I a
subject discharged the fire bottle before tie warning thus
preventing the temperature fom increasing into flie dert
range and avaling a fire warning he did rot technically
have to divert but prudence recommended diverting anyhow
because of the strong pssibility of fre.
Three of the faults required subjects to descend (1) the
cabin altitude inaease^ and after engine irutdown Sx both
(2) the EGT increase and (3) the oil qiantity decrease. Also,
checking weather Si the diversion airport, telling the flight
attendant what was happening and calling dispatch to let
the company kiow lie current situation was not ecplidtly
required but was cnnsidered good airmanship. Subjects
were n of penalized in tie data analysis if they did rot
perform these xtions.
The six NASA-TLX individual workload ratings — mental,
physical and tempcral (tmand, performance, effort, and
frustration — were rormaHzed on ascale fom to 100 with
as low workload and 100 as high workload. They were
combined into an a/era^ workload rating Sx each subject
by data run. These average normalized workload Btin^
were then analyzed using the analysis of variance procedure
in SPSS [14].
In the analysis of the data, significance for both p and x )
was taken at the 005 fevel. Also, for main-orda" effects, a
Tukey HSD post hoc test was done [15].
RESULTS
Benefits of IVedidive Information
If a subject did rothing at all, an alert would occur during
flight fir the 1-, 5-, and 15-minute initial times to an dert.
Subjects could affect lie time to an dert for Jhe EGT
inaease and forward cargo cverheat fiults, or trey could
prevent the parameter fom Baching an alert range altogether
but, in all cases, they had to actively confront the failure.
For t he initial time to an dert of ErA+45 minutes,
subjects did not have to do anything ance an alert would
not be Eached until after knding.
Alert O (curroice: Fcr t he 1-, 5-, aid 15-minute
initial times to ai alat, virether or not ai alert occurred
depended on tie initial time to ai alert ^ sO.Ol). Out of a
possible 48 alerts, only 19 occurred ^able 2). As seen in
table 2, the greater the initial time to an dert, the more
often subjects avoided an alert. Hence^ subjects were taking
actions to lessen the sverity of the iailure, to lessen tie
time pressure ^sociated with the alert, and to lessen its
consequaices.
Table 2 -Numba' of Alert Occurrences
I Alat
Initial Time to Alert ^inutes)
Present
1
5
15
ETA+45
Total
Divert
Dedare
Emo-gency
Yes
No
Yes
No
11
2
8
7
4
5
9
2
8
1
16
17
15
12
Yes
No
10
2
8
4
1
11
12
19
29
Action Before or ^ter Alert Since subjects had time
to deal with the fiiluie before ai alert, vJietha" they acted
before or after an aleit occurred was of interest. In both the
direction and time conditions, subjects brought up the
checklist before aiy alerts ()(fi0.02) (table 3). This is not
surprising b ecause in the direction and time oonditims,
subjects had foreknowledge of the alert and whidi diecklist
was pertinent. F urthermore, the mmber of checklists
accessed before an alert in the baseline condition m^ be
artificially high because subjects were primed fir a feilure.
Thus, they may have been more diligent in canning the
instrumaits boking for deviations.
Table 3 -Number of Checklists Retrieved for the
1 -, 5-, and 15-Minute hitial Times to an Alert
Before Alert
Predictive Information
Total
Baseline Directim Time
Yes
No
8 12 12
4
32
4
Predictive information was dso agnificant in cfetermining
whether pilots accEssed dieddists for the ETA+45 minute
conditioi (xfi0.02) (table 4). No subjects in the baseline
conditio! accessed a diecklist but subjects did when they
had direction or lime information Again, 1his was because
they had an dvisoy message telling 1hem \4iich diecklist
was relevant to the Mure.
Table 4 -Numba' of Checklists Retrieved for
ETA+45 Mnute Initial Time to an Alert
Retrieved
Predictive Information
Baseline Directioi Time Tota
1
Yes
No
4 3 7
4 15
As the initial time to an alert increased ip to the 15-minute
conditioi, the number cf subjects divffting and declaring
emffgendes before an dert range was reached also increased
(XsO.Ol for both) (table 5). S ince they knew the
information to be xcuiate, subjects (fecided to confiont tie
problem before the alert; the more time they had befoe an
alert, ftie moe likely Ihey would declare ai emffgency (in
order to get preferential handling Ifom ATC) and divert
before the alert. In three rases, subjects did rot divert if the
alat was going to occur during tie flight. These three cases
involved the cargo fire failure aid flie subjects disdiarged
the forward cargo fire bottle before ai alert was reached. Qn
the other hand, one subject in the ETA+45 ninute direction
conditioi did d ivat for the oil quantity cfecrease iailure
even tiiough it was not required.
Table 5 -Number of Diveisions and Emergencies
Before Aert
Initial Time to an
Alert (minutes)
1 5 15 Total
Woriiload: For wodcload, predictive information and
initial time to alert were agnificant, ps0.04 for both. As
expected, the ETA+45 minute initial time to an dert had
the 1 owest wxkload rating (able 6). This was because
subjects did not have to confront the feilure, ff they even
noticed the problem, since tie alert was gjing to occur after
landing.
Table 6 -Workload Ratings
Factor
Mean
StDev
Initial Time 1
41
19
to Alert 5
41
19
(minutes) 1 5
41
9
ETA+45
26
13
Predictive Baseline
28
15
Information Directioi
43
18
Time
40
18
Note: 0= low workload, 100= hi^ workload
For p redictive information, workload was rated
significantly 1 owff for the baseline cnndition tian tie
direction c ondition (ti)le 6). The b aseline predictive
information was fimiliar to the aibjeds since tiis b flie
information they currently ibc and this contributed to its
low workload rating.
Unlike the time condition, aibjects had to estimate haw
mucii time they had befoe an alert range would be reached
for the direction predictive information. The oily w^ to do
this was to qjproximate the parameter's rate ofchan^. This
appeared to hcrease the workload
The greatest contributo to workload spears to be dioosing
which adions to carry out. No procedures were gven
reading t he use of direction and time predictive
information and this, most ikely, accounted for subjects
rating workload io time predictive information dosa to
direction predictive i rformation than to the baseline
conditioi. Apparently, deciding oi flie proper course of
actions for tie direction and time conditions increased
workload more than estimating flie time to ai dot lo the
direction condition.
Predictive Information Type and IVediction Times
As mentioned earlier, subjeds could take sveral actions
during fflch frilure ranging from trying to affed flie cmsd of
an alert to dverting to an alternate airfield. As expected, the
predictive information available and the initial time to ai
alat affected wiiai subjects initiated a particular xtion
during tie 1-, 5-, and 15-minute initial times to an alert
The time of checklist a;cess was heavfly influenced ty the
availability of predictive information ^sO.Ol) (table 7). The
baseline condition was statistically hter flian the drection
and t ime conditions. As ecplained above, flis was not
surprising since s ubjeds knew which checklists were
pertinent before tie alert occurred in these two conditions.
Table 7 -Normalized Checklist Access Time (minutes)
Predictive Mormation
Mean St Dev
Baseline
Direction
Time
0.87 0.54
0.12 0.11
0.10 0.16
With regard t) descending, dverting, checking \^eatha",
dedaring an anergency, calling tie flight attendant, and
calling dispatch, the initial time to an alert was significant
(ps0.03 lor all) (table 8). In aU cases, tie 5- and 15-minute
conditiois w ffe statistically eadier than the 1-minute
conditioi. Basically, tie more time subjects had befcre an
alat, the earlier they performed the actions idative to the
time to an alert.
Table 8 -Normalized Times (minutes)
Initial Time
St
Action
to an Alert
N
Mean
Dev
Descend
1 ninute
7
2.16
1.26
5 ninutes
9
0.87
0.40
1 5 minutes
9
0.46
0.34
Divert
1 ninute
11
4.19
2.15
5 ninutes
11
1.10
0.64
1 5 minutes
11
0.56
0.37
Check
1 ninute
6
5.28
3.87
Weather
5 ninutes
7
1.63
1.19
1 5 minutes
8
0.55
0.55
Dedare
1 ninute
10
2.81
1.63
Emergency
5 ninutes
8
0.95
0.53
1 5 minutes
9
0.56
0.35
CaU Flight
1 ninute
10
4.48
2.24
Attendant
5 ninutes
11
1.25
0.62
1 5 minutes
9
0.38
0.28
Call
1 ninute
9
6.m
2.54
Dispatch
5 ninutes
9
1.54
0.80
1 5 minutes
10
0.55
0.33
Since the above results held for all actions, the 1-minute
conditioi was s giarated out ffom the 5- and 15-minute
conditiois because the 1-minute (ondition dd not allow for
mudi time to prepare fa the alert whereas the ether two
conditiois did. The analysis was then redone using the data
from the 5- aid 15-minute initial times to an alert
With the reanalysis for the 5- and 154ninute conditiois,
descent time, d ivffsion time^ and time to cfeclare an
emffgency w ere found to be (tpendent en predictive
information ( table 9). In a 11 cases, lime predidive
information w as significantly lower fiom baseline.
Depending on the action, directicm predictive information
may or may not be diffffent Ifom the baseline condition o
the time condition (table 9). Hence^ the direction predictive
information does decrease the time of when a subject
performs a catain action but this deaease in ime b not as
diiferentiable from the basdine condition as is the time
predictive information.
For all xtions described above, the Sminute condition had
significantly 1 atff times than the 154ninute oonditioi.
Furthermore, even though previous research suggested this,
no interaction occurred between predictive information and
initial time to an alert. In feet, time predictive information
always had an earlier action initiaticxi time. Thus, dthough
pilots reported wanting the diredion type of predictive
information for times to an alert of 5 minutes a less, in
practice, time predictive information appears also to have
the greatest benefits fir the 5-minute initial ime to an alert
Table 9 -Normalized Times for 5- aid 15-Minute
Initial Times to en Alat (minutes)
Predictive
St
Action
Information
N
Mean
Dev
Descend
B iseline
12
0.94
0.41
D recti
12
0.77
0.39
Time
12
0.44
0.33
Divert
laselme
13
1.18
0.60
frection
13
0.84
0.62
Time
14
0.81
0.25
Dedare
Baseline
12
1.04
0.39
Emergency
rectioi
7
0.69
0.31
me
12
0.38
0.24
Note: [= statistical groupings
In general, for a short time to ai alat (1 minute), subjects
did not have much time to use the advance rotificaticsi. For
longer times to an alert (5 and 15 minutes), subjects had
time to affed the timing and occurrence of the alert and to
prqjare for t he alert. In fict, tie time and directioi
predictive i rformation aided them in accessing tie
appropriate checklist, declaring an anergency, aid diverting
before ai alat occurred.
DISCUSSION
To identify t he benefits of jredictive infermation, to
determine the form of predidive informaticai and lan^ of
prediction times, and to corroboiate previous research, a
simulato experimait t esting three types of predictive
information aid four initial times to an alert \^as conducted.
The three types of predictive information were (I) baseline,
(2) direction, and (3) time, and the fiur hitial times to an
alat were (1) 1 ninute, (2) 5 minutes, (3) 15 ninutes, and
(4) ETA+45 minutes. These lactors were found to affed
when s ubjects accomplished (Ertain actions, such as
accessing p ertinent checklists, (tclaring anergaicies,
diverting, and calling the fight attendant and dispatch.
Knowing the remaining time to an alert seemed to produce
the most benefits. For instance, the more ime subjeds had
to deal with the fiilure, the more often they avoided getting
an alert by performing some acticxi such as descending
shutting d own the affeded aigine, o discharging fire
bottles. The initial time to an dert also affeded vhen
subjects performed certain actions. As the initial time to an
alat i rcreased, subjects were moe likely to declare an
emergency and to dvert before a parameter reached an alert
range.
Predictive i rformation also affeded vJien subjeds were
more 1 ikely to access tie appropriate diecklist. With
direction o r time predictive infermation available, tiey
often accessed checklists before an alert occurred.
Finally, the diredion and ime predidive information had
hi^er workload a ssociated wth it than the basdine
information. This was most ikely because subjects had to
decide haw to use the rew infonnation.
Reading the predictiai time, subjects diverted, checked
weather, d eclared an emergency, aid called the fight
attendant and dispatch earlier for the 5- aid 15-minute
initial times to ai alat than for the 1-minute initial time to
an alert
Within the 5- and 15-minute initial times to an alert,
descent time, d ivffsion time^ and time to cfeclaie an
emffgency weie less for the time and drection predictive
information t han tiey \\ere fir the baseline condition,
although the direction condition was not always diffffent
from the basdine condition. This might have beai due to
the heightened a wareness of the subjects to possible
failures. Also, within these two initial times to an alert,
descent time, division time^ time to check weather, time
to declare an emeigency, and time to rail fight attaidant for
the 5 -minute condition were statistically more fcan the
times fcx" the 15-minute condition Lastly, although pilots
indicated i n previous Eseatch an intaaction between
predictive i tformation and initial time to an alert, in
practice, there appears to be no such interaction.
CONCLUSION
The data do suggest that predictive information may be
beneficial to increasing the safety of flight althou^, in this
experimait, the initial time to ai alert more heavily affected
the p erformance. This, most likely, was due to subjects
being primed for failures, S3 they were more actively
scanning the instruments for these Mures. In any rase,
providing the time to ai alert for the longer initial times to
an alert allowed aibjects to prepare fir checklists aid to
dedare emer^ncies earlier in oder to receive preferential
handling from ATC so that they could cfescend and divert
more easily aid timely. Subjects also let ethers know of
the situation e arlier, such as dispatch and the flight
attendants, when they had the predictive information. But
for workload to decrease to the bvel it currentiy is with no
predictive information, flight oews need to become fimiliar
with and folly understand this information.
Other aspects must also be investigated before the fill
usefulness o f predictive information ran be understood.
Further research into the optimal prediction time, axepttble
false alarm rate, and accuracy of the predictive infcamation
must be done. Also, it would be cf benefit to ascertain how
useful the information would be \4ien pilots are not primed
for a failure On the more cperational side, the ability to
estimate the time to an alert with the lalse alarm rate and
accuracy required by the pilots reeds to be investigated
before procedures are cfeveloped ising tiie lime to an alert
predictive information.
REFERENCES
[1] Aircraft A ccident Report, Eastem Airlines, Inc.,
Lockheed L -1011, N534EA, Miami Intemational Aiiport,
Miami, F lorida, M^ 5, 1983, NrSB-AAR-84-04,
National T ransportation Safety Board, Mar. 1984.
(Available fiom NTIS ss PB84 910 404.)
[2] Batelle, Aviation Safety Bq)orting System, Search
Request Numbff 1 929, Aircrafi Subsystems Failures,
Mountain View, CaHforrda, 1990.
[3] E.J. McCormidc and MS. Sanders, Human Factors
in Engineering and Design, Fifth ed. , McGraw-Hill, 1982.
[4] CD. Wickens, Engi.neering Psychology and Himan
Performance, Scott, Foresman& Co., 1984.
[5] E.L. Weiner, B.G. Kanki and RL. Helmreich, eds.,
Codqjit Resource Minagenient, A;ademic Press, 1993.
[6] A.C. Tng'illo, Effects of Historical aid Predictive
Information on Ability of Transport Pilot to Predict an
Alert, N\SA TM-4547, 1994.
[7] A.C. Trujillo, Airline Transport Pilot P references
for Predictive hformation, N\SA TVI-4702, 1996.
[8] 767 F light Manual - Handbook, Revision #72,
United Airlines SFOEG - New Aircraft aid Operational
Engineering, Oct. 1986.
[9] CorrespoidencE with R ichard Gifford, LAL Pilot,
Ret, 1996.
[10] CorrespaidencE w itii Dave Simmon, UAL Riot,
Ret, 1996.
[11] T.W. Wild, Transport C ategory Aircraft Systems,
L\P, fee, 1990.
[12] 757 Q uick Reference Handbook, T he Boeing
Company, Feb. 1992.
[13] S.G. Hart and L.E. Staveland, "Development of
NASA-TLX (Task Load Index): Results of Bnpirical and
Theoretical Research," In P. A lincock & N Meshkati
(Eds.), Human Mental Workload, North-Holland: Elsevier,
1988.
[14] SPSS, SPS^ 7.5 for Windows^, S'SS he, 1997.
[15] D.C. \ioym\\. Statistical Mthods for Psychology, 2"
ed., Boston Duxbury Press, 1987.